Comparing three models to estimate transpiration of desert shrubs

Abstract

The role of environmental variables in controlling transpiration (E_c) is an important, but not well-understood, aspect of transpiration modeling in arid desert regions. Taking three dominant desert shrubs, Haloxylon ammodendron, Nitraria tangutorum, and Calligonum mongolicum, as examples, we aim to evaluate the applicability of three transpiration models, i.e. the modified Jarvis-Stewart model (MJS), the simplified process-based model (BTA), and the artificial neural network model (ANN) at different temporal scales. The stem sap flow of each species was monitored using the stem heat balance approach over both the 2014 and 2015 main growing seasons. Concurrent environmental variables were also measured with an automatic weather station. The ANN model generally produced better simulations of E_c than the MJS and BTA models at both hourly and daily scales, indicating its advantage in solving complicated, nonlinear problems between transpiration rate and environmental driving forces. The solar radiation and vapor pressure deficit were crucial variables in modeling E_c for all three species. The performance of the MJS and ANN models was significantly improved by incorporating root-zone soil moisture. We also found that the difference between hourly and daily fitted parameter values was considerable for the MJS and BTA models. Therefore, these models need to be recalibrated when applied at different temporal scales. This study provides insights regarding the application and performance of current transpiration models in arid desert regions, and thus provides a deeper understanding of eco-hydrological processes and sustainable ecosystem management at the study site.